Face Aging CAAE

Requirements

pip install -r requirements.txt

• torchvision 0.4.0
• torch 1.2.0
• UTKFace Aligned&Cropped dataset

Colab Notebook

Training - Colab Notebook

Inference - Colab Notebook - pending

Snapshot

Usage

• git clone or download zip file of this repository
• download Aligned & Cropped version of UTKFace from here
• Install requirements
• execute main.py

python main.py

Results

UTKFace

rows: years of 0 ~ 5, 5 ~ 10, 10 ~ 15, 16 ~ 20, 21 ~ 30, 31 ~ 40, 41 ~ 50, 51 ~ 60, 61 ~ 70, over 70

epoch:50, step:86
EG_L1_loss:0.075875           |G_img_loss:5.226651           
G_tv_loss:0.003358            |Ez_loss:0.851948
D_img:0.998970                |D_reconst:0.015672            |D_loss:0.017007               
D_z:0.435863                  |D_z_prior:0.606904            |Dz_loss:1.133016 

output with 1.7k images trained for 50 epochs

To-do:

• Check the corretness of model
• Write inference code with trained weights
• Release pretrained weights for this repo

Credits

• Age Progression/Regression by Conditional Adversarial Autoencoder
• Face-Aging-CAAE-Pytorch

Other creations

• AgeProgression-Pytorch
• AgeProgression-tensorflow